In recent years, a function of reading aloud text data from mails and website contents using a voice is incorporated into embedded equipment such as cellular phones. In a speech synthesis process for realizing such a read-aloud function using a voice, a waveform dictionary as a database storing speech segment data necessary for synthesized speech by compressing the data with the use of a compression method such as ADPCM (Adaptive Differential Pulse Code Modulation) is preliminary recorded in recording means such as a built-in memory. When generating a synthesized speech waveform, a compressed speech segment data read from the wave function dictionary is expanded and decoded. Then synthesized speech is outputted on the basis of the generated speech signal by performing processes, such as combining the expanded and decoded speech segment data and adjusting the pitch and speed.
According to the Japanese Laid-open Patent Publication No. H08-160991, a speech-segment production method and a speech synthesis method are discussed.
However, the expansion and decoding of a speech signal compressed by a compression method such as ADPCM sometimes cause deterioration in the sound quality of the generated speech, such as noise and non-smoothness. Moreover, deterioration in sound quality, such as noise and non-smoothness, may also occur when combining a plurality of speech segment data and adjusting the pitch and speed of speech.